Hello Ben, In short:
- Sista hotspot detection can be used for other things than the optimizer easily, so you can do what you want with it and it has been stable for a while now. - If you want to optimize your code, you can already use the time profiler, it's likely to be better than the hot spot detection to find out what to optimize and it gives you information about which method call what method. - You can't solve the problem of instruction selection / register allocation from the image with the current interface image-VM, hence a human doing it instead of the optimizer doesn't solve the issue. I've already worked on the JIT back for these problems, there is still some work to do but we are solving at least partially the issue. Cheers Clement 2015-11-01 2:59 GMT+01:00 Ben Coman <[email protected]>: > I just bumped into a very nice read on Sista that I had missed before... > http://www.slideshare.net/esug/sista-talkesug2 > > Slide 52 states cons are... > Some optimisations are difficult: > * Instruction selection > * Instruction scheduling > * Register selection > and just thinking out loud, I wonder what possibility there might be > after Sista that its hotspot detection could be used to > flag/prioritise methods which a human could manually optimise, > providing & recording the conditions under which that human-optimised > method/call-stack can be used ?? > > cheers -ben > >
